GitHub has launched a local stacked pull request workflow via a brand new CLI extension known as gh-stack, closing a spot that third-party instruments have crammed for a number of years. It goals to resolve the issue the place giant pull requests are exhausting to overview, sluggish to merge and liable to conflicts, with GitHub stating that reviewers lose context, suggestions high quality drops, and the entire crew slows down.
Stacked pull requests, generally known as dependent or chained PRs, are a code-review sample by which every department targets not the primary department however the department instantly beneath it in a sequence. The strategy lets builders proceed work on later layers of a function whereas earlier layers are nonetheless underneath overview.
“I used to be spending a variety of time ready for code overview to occur, which was a serious motivator for constructing the software.”
Evan Priestley, co-creator of Phabricator Differential
Analysis cited within the launch announcement suggests the workflow has measurable advantages. An evaluation of 1.5 million pull requests discovered that PRs between 200 and 400 traces had 40% fewer defects and have been permitted 3 times sooner than bigger ones. The stacked strategy is designed to maintain particular person PRs inside that vary even when the underlying function is substantial.
The gh-stack extension integrates with the present GitHub CLI and handles the mechanics which have traditionally made stacked workflows painful to keep up manually. The first command, gh stack sync, cascades a rebase throughout your complete stack and force-pushes every department atomically after a change to an earlier layer. GitHub’s pull request interface beneficial properties a stack map that lets reviewers navigate between layers, and department safety guidelines are enforced in opposition to the ultimate goal department reasonably than the quick base of every PR. CI equally runs as if every PR have been focusing on the primary department instantly.
The extension additionally ships with an AI agent integration. Working gh ability set up github/gh-stack teaches suitable AI coding brokers the right way to create and handle stacks, permitting them to interrupt a big diff into layers or develop with stacks from the start of a process.
“The CLI is totally non-compulsory, you may create stacked PRs purely by way of the UI.”
Sameen Karim, GitHub
The stacked diff sample just isn’t new to software program improvement. Meta and Google each adopted comparable workflows almost a decade in the past, constructing customized tooling, together with Phabricator and Gerrit. Simohamed Marhraoui wrote in LogRocket as early as 2021 that the strategy was already relevant to GitHub, although it required care round merge technique: squash and rebase merges each rewrite commit hashes, breaking the id monitoring that hyperlinks branches in a stack. The restriction Marhraoui famous, that solely a typical merge commit ought to be used for intermediate PRs in a sequence, stays a sensible concern with any stacked workflow on GitHub.
Alan West, writing on dev.to shortly after the launch, noticed that git itself presents no assist managing the connection between dependent branches. “If you rebase the primary department, each downstream department must be rebased too, manually,” West wrote, describing a sequence of 5 steps that builders should repeat every time a reviewer requests a change on an early PR. West notes that three to 4 PRs is a sensible ceiling for a stack: “Past that, the cognitive overhead of monitoring dependencies begins to outweigh the overview advantages.”
The principle competitor to GitHub, Graphite, based by former Meta engineers, is now accessible and not using a waitlist right this moment. Graphite presents a stack-aware merge queue, a web-based overview interface, VS Code integration, and a CLI. Its free tier contains the CLI and stacking workflow; paid plans begin at $20 per person per 30 days. Joe Buza noted on LinkedIn in February 2026 that he had been prompting AI coding brokers to interrupt options into stacked PRs utilizing a Graphite-style workflow, constraining every PR to underneath 200 traces and requiring that every layer “do one logical factor and make sense by itself.”
Neighborhood response to the GitHub announcement has been blended. The Hacker Information thread on the launch reached 516 factors and 282 feedback. One strand of the dialogue welcomed the transfer as mainstream validation of a sample that had been the protect of huge engineering organisations: “Stacked diffs have existed at Meta for a decade, glad GitHub is becoming a member of 2016.” A sceptical strand questioned whether or not the workflow is sensible in any respect on GitHub’s infrastructure: “Both modifications are unbiased and you utilize separate PRs, or they’re dependent and reviewing them individually is unnecessary.” An additional line of criticism, summarised by ByteIota, pointed to squash-merge compatibility and cascading rebase conflicts as unresolved technical issues that Graphite has had years to work via.
GitHub’s entry into stacked PRs is notable for one function that third-party instruments can not provide: the stack map and enforcement logic stay contained in the pull request UI itself, which means reviewers don’t want a separate account or extension to see the context. The official documentation notes that the CLI is “fully non-compulsory” and that stacks may be created via the GitHub UI or API instantly. Whether or not that native integration is sufficient to appeal to groups that already use Graphite, or convey the workflow to builders who’ve by no means tried stacking in any respect, will rely largely on how GitHub handles the sting circumstances in the course of the non-public preview interval.
The function entered non-public preview on 13 April 2026, and builders should be part of a waitlist earlier than the extension will operate for his or her repository. Associated protection on InfoQ features a February 2026 piece on GitHub’s reworked layered defences for infrastructure coverage enforcement, and an April 2026 article on Anthropic’s agent-based code overview for Claude Code, which discovered that substantive overview feedback on PRs with greater than 1,000 traces modified rose from 16% to 84% after adoption of automated overview tooling.